ExBox: Experience Management Middlebox for Wireless Networks
by Ayon Chakraborty (Stony Brook University), Shruti Sanadhya (Hewlett Packard Laboratories), Samir R. Das (Stony Brook University), Dongho Kim (Hewlett Packard Laboratories), and Kyu-Han Kim (Hewlett Packard Laboratories)
The networking environment is fragmented across diverse applications, different devices and variety of networks. A key question is how to ensure good Quality of Experience (QoE) in this fragmented environment. Moreover, QoE mean different things depending on the application. For web browsing, QoE can be measured using the page load time, whereas for storage, it can be measured using file syncing time. This paper aims to model network capacity in terms of QoE, which they call as "experiental capacity".
The paper uses a data-driven approach in which they observe the QoE values for different mixes of network traffic. They use a regression model to then determine how different network flows affect the QoE of applications. Finally, the authors compare this technique with some baseline techniques based on Cisco's routers and show that it has better precision and accuracy.
Q: Did you look at subjective metrics of QoE?
A: No, we have only looked at objective metrics.
Q: Did you look at diversity within applications?
A: No, but a more sophisticated regression model can handle it.
Q: Did you look at buffering ratio or other things in video?
A: No, but that was mainly because no buffering delays were observed when the videos were played. But our framework is general and can be applied to other QoE metrics such as buffering ratio too.
Q: What about including other factors such as mobility and SNR?
A: While some other factors can be included, current model does not allow including too many features since it would lead to explosion of feature space.
AmorFi: Amorphous WiFi Networks for High-density Deployments
Ramanujan K Sheshadri (University at Buffalo, SUNY), Mustafa Y. Arslan Karthikeyan Sundaresan (NEC Laboratories America), Sampath Rangarajan (NEC Laboratories America), and Dimitrios Koutsonikolas (University at Buffalo, SUNY)
Providing WiFi has become very important in crowded spaces such as stadiums and airports. Current state-of-the-art deployment solutions estimate the per-user bandwidth requirements by determining the aggregate throughput for a particular area. This is then used to decide the number of Access Points (APs). However, this technique has a number of disadvantages. First, it requires overprovisioning of resources and secondly, it cannot accomodate broadcast traffic. Broadcast traffic usually requires more area coverage and much bigger cells. Thus, a more flexible WLAN architecture that can provide capacity and coverage is needed.
In this paper, the authors propose to utilize Cloud-RAN architecture for WLANs. In other words, they propose to decouple radio frequency transmissions from baseband processing. The baseband processing is done separately in a data center, while the RF transmission is done by remote radio heads (RRHs). This increases flexible reconfiguration of the network depending on demand by altering the mapping of APs to RRhs.
The actual mapping is done by mapping it to a graph-coloring problem, and then utilizing an iterative solution. Various experiments have shown that the throughput can be doubled compared to traditional WLANs.
Q: One appealing benefit of C-RAN is saving electricity, since it leads to 3-50% reduction. For wireless LANs isn't this benefit reduced?
A: Since we target high-density locations, there is a significant cost to deploy the APs -- electricity and manual labor. Compared to that, the overhead cost of implementing this technique is not that high.
Q: Have you calculated the bandwidth and latency between the RAU and BBU?
A: Bandwidth is not a big issue. Latency of optical fibres is not that significant problem. The DIFS time is 802.11n is 32-36 microseconds, which is more than what the optical fibres deliver.
Q: Channel capacity of RRH is also important. Isn't it?
A: RRHs are just antennas, and thus their capacities are rarely a bottleneck.
SlickFi: A Service Differentiation Scheme for WiFi
Kamran Nishat (SBA School of Science and Engineering, LUMS), Farrukh Javed (SBA School of Science and Engineering, LUMS), Saim Salman (SBA School of Science and Engineering, LUMS), Nofel Yaseen (SBA School of Science and Engineering, LUMS), Ans Fida (SBA School of Science and Engineering, LUMS), and Ihsan Ayyub Qazi (SBA School of Science and Engineering, LUMS)
Recent WLANs carry a variety of traffic with real-time constraints (for audio or video) and bulk transfer (for cloud storage).
Recent WiFi standards support more than 1 Gbps throughput by allowing aggregated frames and block acknowledgments.
However, aggregation and block acknowledgments also increase delay for real-time applications.
This is clearly demonstrated using some iperf-based experiments.
In this work, the authors propose adapting the channel depending on traffic to both handle real-time traffic and allow bulk transfer.
They implement this technique by making driver-level changes to their laptop.
By deploying it in real environment, they show that it can provide better-quality videos and also provide better throughput.
Q: There's an article in LWN "Make WiFi fast". Can you combine that technique with this?
A: I will definitly look into it.
Q: For communication between different radios, how fast can you change from 20 MhZ to 40Mhz?
A: Its already built into the hardware and happens into the microsecond level granularity. Since packet latencies are in milliseconds, this is not a major factor.